Re: [patch] for PR 19333

[Zdenek Dvorak <rakdver at atrey dot karlin dot mff dot cuni dot cz>]

Hello,

> On Mon, Jan 10, 2005 at 09:50:15AM +0100, Zdenek Dvorak wrote:
> > extern short ac[][];
> 
> This makes no sense.  Surely it's invalid C?
> 
> Certainly *something* is missing in the front end, either via a
> diagnostic, or a cast.  I cannot think of any reason you should
> have to deal with dereferencing incomplete types here.  Or *was*
> there a cast and it got lost somehow?

as for validity of C -- I don't know.  Gcc accepts it, however.

Note that we do not dereference incomplete type anywhere in the example --
everything is always casted to a suitable type.  Assignment ac.33 = &ac
does not seem to need a cast, since (short [] *) and (short[][]) are
considered equivalent by STRIP_USELESS_TYPE_CONVERSION.  (*ac.33)[0] is
fine - type of *ac.33 is (short []) and there is no problem with
indexing this.  Once constant propagation is performed, you get
&(*&ac)[0], which has the discussed expression (*&ac) as a
subexpression, and it has maybe_fold_offset_to_array_ref called for it
during recursive simplification in fold_stmt_r.

Zdenek